1. Introduction
Personalized products are emerging and have become much more feasible for production with 3D printing technologies. Examples range from print-at-home applications, personalized sports equipment, to jigs and fixtures used in industrial production (Reference Fiedler, Ehrenstein, Höltgen, Blondrath, Schäper, Göppert and SchmittFiedler et al., 2024; Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). Mass personalization is characterized by involving customers in the early development process (Reference Poot, Wehlin, Tarkian and ÖlvanderPoot et al., 2020; Reference Wang, Ma, Yang and WangWang et al., 2017). However, many customers do not possess the design knowledge, the CAD modelling expertise and the manufacturing knowledge that are required to come up with suitable designs on their own (Reference Fuchs, Bartz, Kuschmitz and VietorFuchs et al., 2022; Reference Wiberg, Persson and ÖlvanderWiberg et al., 2019). Other customers, particularly in industrial contexts, may possess these skills but would still benefit from the elimination of repetitive and manual design tasks. Therefore, there is a need for tools that enable users of varying skill levels to generate production-ready parts with minimal effort (Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019).
Configurators address this issue by mapping user requirements to product configurations. They do so by taking user input and generating a design from it according to a given framework. This is strongly related to the area of knowledge-based engineering (KBE), which addresses methods, processes and tools intended to capture and encode engineering knowledge to automate repetitive engineering tasks (Reference Kügler, Dworschak, Schleich and WartzackKügler et al., 2023; Reference Verhagen, Bermell-Garcia, Van Dijk and CurranVerhagen et al., 2012). In this context, design automation (DA) refers to systems which encode such engineering information to automate specific repetitive and manual activities. DA has the potential to significantly reduce time and cost within the design process by quickly providing the user with updated designs for varying inputs (Reference Kügler, Dworschak, Schleich and WartzackKügler et al., 2023; Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019; Reference Verhagen, Bermell-Garcia, Van Dijk and CurranVerhagen et al., 2012; Reference Vidner, Wehlin and WibergVidner et al., 2022). The configurator acts as an interface between the customer and the manufacturer, making it a critical element of mass personalization by serving as the main communication tool in the customization process (Reference Chen and TsengChen & Tseng, 2007; Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019; Reference Poot, Wehlin, Tarkian and ÖlvanderPoot et al., 2020). It allows for co-creation or co-design, where both the customer and the manufacturer negotiate the details of the product (Reference Chen and TsengChen & Tseng, 2007; Reference KumarKumar, 2007). Configurators are based on a predefined DA framework with fixed functions, rules, and components, spanning the intended design space of the configurator (Reference Poot, Wehlin, Tarkian and ÖlvanderPoot et al., 2020). This predefined design space limits the degree of achievable personalization, making configurators struggle to accommodate unforeseen customer requirements (Reference Poot, Wehlin, Tarkian and ÖlvanderPoot et al., 2020). The definition of an appropriate DA framework therefore becomes a critical aspect of configurator development. The development and improvement of such a DA framework requires its implementation in concrete tools to identify practical challenges (Reference Vidner, Wehlin and WibergVidner et al., 2022). However, existing implementations of DA tools are often highly specialized, which limits their reuse and hinders broad evaluation (Reference Umland, Winkler and InkermannUmland et al., 2023).
A key research gap therefore lies in the lack of flexible DA frameworks that have been evaluated through extensive user feedback. In this work, we investigate how DA frameworks can be meaningfully evaluated through a deliberately simple, real-life implementation. We first present a flexible and modular DA framework that can be extended and adapted to various applications. The proposed evaluation approach is demonstrated by implementing the framework in a proof-of-concept application: the Bike Connector Tool, which serves as an exemplary test system for the framework. The tool addresses the simple yet useful application of attaching accessories to bicycles. This enables participation of users with varying backgrounds and technical expertise. The application also covers a wide variety of possible parts, since geometries of bike components and accessories can vary greatly, as can be confirmed by looking at the broad range of commercial connectors available. The resulting parts are small, lightweight, non-safety critical, and can be manufactured using consumer-grade material extrusion (MEX) additive manufacturing (AM), allowing rapid home-based fabrication and testing of generated designs. A small case study is used to demonstrate how meaningful insights about the DA framework can be derived via the simple, real-world implementation. Finally, the feedback and learnings from the case study are discussed and related to the underlying framework.
The contribution of this paper is an evaluation approach for DA frameworks based on deliberately simple and accessible implementations that enable practical end-to-end assessment and broad user feedback.
2. Related works
A popular approach to mass personalization is parametric modelling (Reference Camba, Contero and CompanyCamba et al., 2016). Reference Hafemann, Daumoser and LienkampHafemann et al. (2023) propose a parametric model that can visualize the exterior design of autonomous shuttles. The model has been developed for use in early phases of shuttle design. It serves as a convenient tool to show the trade-offs between functional considerations stemming from sensor integration and exterior design considerations. The implementation has been evaluated by its ability to model three existing reference shuttles. The designs in this case are adjusted by adjusting the model parameters directly. Such an approach might not always be desirable, especially in cases where the users are non-experts. Such users might not be able to translate their needs into discrete parameter values (Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). For these cases, Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al. (2019) propose a mapping of user needs to design parameters. Their case study on personalized kayaks implements an interface that lets users configure their kayak based on criteria such as tracking and manoeuvrability. The proposed method has been validated through qualitative feedback from a professional stakeholder (Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). While the kayaks are intended for AM, validation of manufacturing has been left open for future works (Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). For the case of kayaks, the relationship between customer needs and design parameters follows a logical mapping (Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). However, the authors acknowledge that this might not always be the case. Reference Ozdemir and CasciniOzdemir and Cascini (2020) address one instance where the mapping of user needs to relevant design variables is not straightforward. They tackle this challenge in their case study on personalized saxophone mouthpieces. Here, the mapping from user requirements concerning sound characteristics and playing conditions to design variables has been established experimentally. The parametric design can then be personalized through a dynamic product configurator (Reference Ozdemir and CasciniOzdemir & Cascini, 2020). The dynamic nature of the configurator adapts the valid ranges of successive user inputs such that new inputs do not interfere with the objectives of previous steps. This approach is necessary due to the dependencies between different user requirements and design variables of the parametric model. Evaluation has been performed in a subsequent user study with 5 musicians (Reference Ozdemir, Pàmies-Vilà and VerlindenOzdemir et al., 2022). The personalized mouthpieces have been additively manufactured and compared to reference mouthpieces. Results showed that 4 out of 5 users preferred the personalized mouthpiece, demonstrating the successful mapping of user requirements to custom products. To demonstrate the generality of the proposed method, Reference Ozdemir, Cascini and VerlindenOzdemir et al. (2020) performed an additional case study on knitted footwear, where the parameters influencing knitting design and fit have been identified along with their dependencies.
Instead of explicitly mapping requirements to design variables, optimal configurations of parametric models can be generated using generative design. Reference Kosec, Huic, Martinec and ŠkecKosec et al. (2025) combine a parametric model with multi-objective optimization to generate an optimal design of a dental implant abutment tailored to a sample patient. Functional designs are ensured by embedding constraints into the parametric model. Their implementation has been evaluated on the design of one patient-specific dental implant abutment without proceeding to manufacture the part. They conclude that mass personalization should rely on generative approaches not entirely but rather implement it for the automation of specific design aspects. Reference Poot, Wehlin, Tarkian and ÖlvanderPoot et al. (2020) propose a design framework for personalized spiral staircases. Their dual configuration system combines sales and design aspects of product configuration. The approach is designed to include customers early in the design stage. Optimization is used to propose multiple pareto-optimal variants depending on user requirements such that users can decide on a variant that best satisfies their objectives. Proof-of-concept implementation of the framework showed significant reductions in lead times and the ability to robustly handle even complex cases. While generated designs were evaluated with the stair manufacturer, no physical systems have been manufactured. An approach that gets rid of manual specification of user requirements completely is demonstrated by Reference Li, Ploumpis, Zafeiriou and MyantLi et al. (2020). Their approach generates personalized ventilation masks based on facial topology. A facial image serves as the sole input to the process, from which a trained model reconstructs the face topology and adjusts the parametric model of the mask for optimal fit. The approach has been used to generate 17 custom masks while the production of physical prototypes has been reserved for future works. A more general approach to the design of wearables is given by Reference Rosen, Choi and SinhaRosen et al. (2024). They propose a design methodology that can be used to configure personalized wearables starting from a design library. For each requirement, a library element is selected and adjusted to the relevant body part, which has been demonstrated by generating sketches of two solution principles for a passive exosuit.
The works presented in this section focus on generating practical outcomes through real-world implementations of proposed frameworks. Evaluating these implementations enables the derivation of theoretical insights into the employed approaches and underlying theories (Reference Barata, Cardoso and CunhaBarata et al., 2023; Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al., 2019). Such action research is often performed in cycles consisting of diagnosing, action planning, action taking, evaluating, and specifying learning (Reference Davison, Martinsons and KockDavison et al., 2004). Across the presented cases, the evaluation phase is addressed very differently, with most implementations not including manufacturing and testing of the generated designs. This limits opportunities for learning and iterative improvement. The reasons for limited evaluation vary between cases. Some target complex products, such as shuttles, exoskeletons, or staircases, which are not feasible to manufacture just for evaluation purposes. Even seemingly less complex products can face manufacturing challenges, like AM of large-scale parts in the case of customized kayaks. Furthermore, evaluation through user testing can be challenging, as is the case for medical products such as dental abutments and CPAP masks, or for products targeting niche user groups, such as personalized saxophone mouthpieces. These observations highlight the need for more accessible applications targeting broader user groups, which facilitate evaluation in an end-to-end manner, from user specification to the use of the physical product.
3. Methods
The following sections present the proposed evaluation approach for DA frameworks based on a deliberately simple, accessible, end-to-end implementation (Figure 1). First, the DA framework to be evaluated is described at a conceptual level, without addressing implementation details. Then, the Bike Connector Tool is proposed as a simple but practical implementation of the DA framework. Although the DA framework formally concludes with the generation of a CAD model, a realistic end-to-end evaluation requires physical fabrication of the generated designs to enable collection of empirical user feedback. Manufacturing and physical evaluation are therefore included as part of the evaluation approach shown in Figure 1. Based on the Bike Connector Tool, a case study is conducted that generates feedback on both the concrete implementation and the underlying DA framework.
Workflow of the proposed evaluation approach, showing the DA framework, its implementation in the Bike Connector Tool, and the extension to manufacturing and evaluation for an end-to-end assessment. The user specifies the desired part by selecting features from a parametric feature library, configuring their parameters, and defining their spatial relationships. Objectives and constraints may be specified before triggering the automated design generation. Although the framework formally concludes with the generation of a CAD model, the workflow is extended through manufacturing and physical evaluation of the generated part to enable empirical user feedback
Figure 1 Long description
A diagram of the design automation framework for personalized product manufacturing. Panel A: Selection. The user selects features from a parametric feature library. Panel B: Configuration. The user configures the parameters of the selected features. Panel C: Positioning. The user defines the spatial relationships of the features. Panel D: Problem formulation. The user specifies objectives and constraints. Panel E: Connector types. A library of different connector types is shown. Panel F: Design algorithms. The design algorithms optimize the connection and perform smoothing and filleting. Panel G: Additive manufacturing. The CAD model is manufactured using additive manufacturing techniques. Panel H: Physical evaluation. The final part is physically evaluated to enable empirical user feedback.
3.1. Design automation framework
The DA framework evaluated in this study is shown as part of Figure 1. In its most general form, it consists of a parametric feature library and a set of design algorithms. Users interact with the framework by selecting one or more of the parametrized design features relevant to their application from the feature library. Each feature exposes a set of configurable parameters that allows its dimensions and functional properties to be adapted. After configuration, features can be positioned and oriented in space through translation and rotation, enabling the definition of spatial relationships between them. The configured and positioned features are then passed to the design algorithms, which generate a final design based on predefined logic. Depending on the specific algorithm, users may additionally specify objectives and constraints that guide the automated design process. Once the relevant selections, configurations, spatial relationships, objectives, and constraints are defined, the user’s design intent is fully captured. This information is provided to the design algorithms, which process the inputs to generate final designs.
3.2. Implementation: the Bike Connector Tool
In the scope of this work, bike connectors have been defined as parts which serve the purpose of connecting any type of accessory to a bicycle. They consist of two interfaces connected by a connection body (Figure 2a). One interface is attached to the bike, while the other attaches to an accessory such as a (bike) light or computer. The technical implementation of the Bike Connector Tool combines a web-accessible user interface (UI) with the selected DA framework implemented using Rhino Compute. The UI (Figure 2b) guides the user through the personalization process. Designing a bike connector starts with selecting two interfaces from the feature library. The values of the exposed design parameters of these features are then adjusted to match the relevant dimensions of the bike and the accessory, with the preview in the UI updating after each adjustment. The adjustable parameters depend on the type of connector selected but typically allow users to change the overall size, clamping diameter, or amount and size of mounting holes. After configuring the two connectors, their position and orientation with respect to each other can be adjusted. Finally, the user defines the target diameter of the connection body, based on the perceived requirements regarding stiffness and strength. The user is allowed to jump back and forth between these steps, until satisfied with the defined inputs. Finally, the automated design pipeline is triggered. From this point onward, the process is fully automated, with no further opportunities for user intervention. The design algorithms generate the connection body that links the two connector interfaces (Figure 2a). The process is limited to geometry generation and does not include manufacturing or structural considerations. Instead, an iterative procedure is employed to generate a connection body that prioritizes a direct, smooth connection between the interfaces while conforming to the specified target diameter of the connection body. The design space of the connector body is further constrained to avoid interference with surrounding parts to which the connector will eventually be mounted, as well as to avoid interference with the mounting hardware. Smooth transitions between the connector interfaces and the connection body are added prior to post-processing the geometry to improve surface quality. The final geometry is displayed in the UI for inspection and can be exported in the STL format for additive manufacturing.
(a) Connector consisting of two interfaces linked by the connection body and (b) user interface of the Bike Connector Tool
3.3. Evaluation: case study
To evaluate the proposed DA framework as implemented in the Bike Connector Tool, a qualitative case study was conducted with two primary objectives. First, the study seeks to demonstrate that meaningful evaluation of DA frameworks requires consideration of the entire end-to-end process chain, from design to physical realization and use, rather than the design tool alone. Second, it aims to demonstrate that meaningful insights about a DA framework can be obtained by implementing it in a simple, real-world application that can be used by a broad range of participants.
Based on these objectives, the study examined whether participants could successfully configure and generate functional connectors and how their feedback could be used to improve both the tool, as well as the underlying DA framework.
In total, 19 participants took part in the case study. The participants included volunteer students from an engineering course as well as research associates and PhD students from an engineering research group. All participants therefore had a technical background, with varying levels of CAD experience ranging from a basic understanding to expert level.
Participants were given a brief introduction to the tool and were then tasked with designing a connector to attach an accessory of their choice to a bicycle. A selection of accessories, such as lights, inner tubes, and hand pumps, was provided alongside bicycles of different types. Participants were also free to design connectors for their own accessories and personal bikes, with the intention of encouraging creativity and revealing applications not anticipated by the tool’s developers.
All connectors were subsequently 3D-printed in PETG using MEX by a staff member. The printed connectors were then returned to the participants, who were encouraged to evaluate the parts by mounting them on the respective accessories and bicycles.
Feedback was collected using a design log. Participants first documented their design intent, including the accessory type and mounting location. In a second section, they provided general feedback and reflections on tool usability, desired improvements, and how well the outcome matched their intended design.
4. Results and discussion
The following section presents the feedback and insights gathered from the conducted case study and discusses them in the context of the proposed framework of Section 3.1. In total, participants designed 21 connectors, which are shown in Figure 3.
Overview of the 21 connectors created by participants in the case study
All connectors were successfully produced; however, observations during mounting showed that roughly half of the connectors would have required a second or more design iterations to achieve a fully satisfactory result. Both the successful and the unsuccessful cases provide valuable insights. From the results of the case study four key learnings are identified:
-
• Tool perception: the tool enabled intuitive interaction and allowed users to efficiently translate design intent into concrete designs
-
• Library limitations and generic interfaces: limited library coverage constrained some applications, while generic, well-parameterized interfaces emerged as critical enablers for broad applicability
-
• Limits of automation: the framework must allow targeted user intervention when automated outcomes do not align with user expectations
-
• Need for visual aids and spatial reference: insufficient spatial reference made it difficult for users to judge scale, orientation, and fit of their designed part
To illustrate the learnings, four representative connectors are presented below, each accompanied by the associated user feedback and a discussion of the resulting implications for the presented framework.
4.1. Tool perception
Connector C1 for mounting a bicycle computer to the handlebar. From left to right: exported STL, printed connector, mounted connector on the bike. The rightmost panel displays the framework with the elements addressed by this learning marked in red
The case study indicates that the proposed framework and its implementation are perceived as intuitive and easy to use, enabling users to quickly translate an initial idea into a concrete connector.
Connector C1 (Figure 4) illustrates this behaviour. It was designed to mount a bicycle computer centrally in front of the stem using a handlebar clamp and a corresponding device interface. The connector was successfully manufactured and tested, and the participant described the overall process as simple and quick. This feedback reflects a large share of the responses. Most participants found the Bike Connector Tool easy and intuitive to use, stating that “The tool is fast and easy with good results” and characterizing it as “very efficient and easy to understand”. With respect to the outcome, most participants commented that the generated part matched their expectations, noting that the “output matches the idea” or was “exactly what was envisioned”.
The high number of successfully generated and mounted connectors, together with the consistently positive usability feedback, indicates that the proposed DA framework and its implementation in the Bike Connector Tool support an efficient transition from design intent to production-ready geometry with relatively little effort.
4.2. Library limitations and the value of generic interfaces
Connector C2 for mounting spare batteries for electronic shifting to the bottle-cage mounts of a mountain bike. The battery is secured using a compliant clamp interface with a rubber strap. From left to right: exported STL, printed part, mounted connector, and relevance to the framework
The case study shows that the usefulness of the framework is strongly influenced by the scope and quality of the feature library. While participants frequently perceived the feature library as too limited, adaptable generic interfaces proved highly effective in bridging these gaps.
Connector C2 (Figure 5) provides a representative example. It was created to attach spare batteries for an electronic mountain bike drivetrain to the bottle cage mount. Since no dedicated interface for this accessory existed in the feature library, the participant initially perceived the available options as limited and suggested that the tool should support uploading custom interfaces to better cover specialized applications. Similar feedback appeared throughout the case study; with participants noting that “it would be nice to have more interface options”.
This reflects a fundamental challenge for DA systems: depending on the application, the variety of user requirements are very broad, making it difficult to create a predefined feature library that satisfies all needs. Despite the limitations, the participant of C2 achieved a successful design by using a generic rubber-clamp interface from the library. Although not intended specifically for the battery, this adaptable interface could be adjusted to hold it securely and proved equally suitable for many other accessories. Generic connector types such as the rubber clamp or simple extruded features were used repeatedly and creatively across different scenarios.
Generic interfaces, however, require a higher degree of adaptation and, consequently, greater technical understanding on the user side. This can reduce the accessibility of such tools for less experienced users.
For the framework, this suggests a layered approach to feature library design. A set of easy-to-use standard interfaces that require little adaptation should be provided to support non-expert users and lower the entry barrier. In parallel, a set of highly adaptable and broadly applicable interfaces, together with the option to upload and share custom interfaces, should be offered to enable more technically experienced users. An example of such a layered approach can be found in the case study of Reference Lithgow, Morrison, Pexton, Panarotto, Müller, Almefelt and McLarenLithgow et al. (2019), where the UI of a kayak configurator allows to switch between simplified options for beginners and full customization of features for advanced kayakers.
4.3. Limits of automation: option for manual control
Connector C3 for mounting a front light to the handlebar using a bolt plate that attaches to the light via the thread on its underside; from left to right: exported STL, printed part, mounted connector, and relevance to the framework
The case study reveals that fully automated geometry generation does not always align with the outcome that users intuitively expect. Connector C3 (Figure 6) illustrates this limitation. It was designed to attach a front light to the handlebar using a handlebar clamp as the bike-side interface and a plate with a hole as the light-side interface. Although both interfaces could be configured as intended, the automatically generated connection body between them was strongly curved and appeared unintuitive to the participant, despite being technically valid. Similar observations were reported in other cases, where users described the connection body as “too curved” or stating that it “does match the idea to some extent but not as sleek as it could have been”. Several participants explicitly expressed a desire to manually adjust the connector path or otherwise influence the algorithm responsible for generating the connection body. These cases illustrate a central tension in DA: automated algorithms can generate valid geometries, yet the results do not always match what users intuitively expect. Connector C3 shows that, even when most of the design work is automated, users may still recognize when the generated geometry deviates from what they consider reasonable. For the framework this insight suggests that, even in a tool like the Bike Connector Tool, which aims to relieve users of as much design work as possible, there should be an option to manually influence automated steps whenever the algorithm produces a result that is perceived as unsatisfactory.
4.4. Need for visual aids and spatial reference
Connector C4 for mounting a front light to the handlebar using a two-piece clamp and a through bolt attachment for the light; from left to right: exported STL, printed part, mounted connector, and relevance to the framework
Figure 7 Long description
Panel A: A 3D model of a connector for mounting a front light to a bicycle handlebar. Panel B: A 3D-printed version of the connector. Panel C: The connector mounted on a bicycle handlebar. Panel D: A diagram of a design automation framework showing the steps from user input to digital manufacturing.
The case study shows that insufficient spatial reference in the user interface makes it difficult for users to correctly assess position, orientation, and scale when designing connectors for objects, often leading to misjudged dimensions and alignment issues. Connector C4, shown in Figure 7, also mounts a front light to the handlebar, but in contrast to C3, the light is positioned underneath the handlebar and centred along the bike’s longitudinal axis. The participant reported that the most challenging part of the design process was imagining the relative position of the light and the handlebar while personalizing the connector. In the current implementation, the UI offers a 3D preview of the connector but does not provide any explicit spatial reference elements, such as a handlebar or bicycle model. This lack of context was reflected in several comments, where participants noted that “It’s challenging to imagine how it would look on the bike” and that it “would be helpful if there is a bike showing how the tool would look on the bike”. These difficulties contributed to misjudged dimensions and orientations, which in some cases led to misalignment, poor fit, or unexpected visual appearance once the parts were mounted on the actual bicycle. As a result, this constituted the primary reason why most connectors would have required a second design iteration to meet user expectations. Because these issues often only became apparent when participants attempted to mount the physical part to the bike, this underscores the importance of evaluating DA frameworks beyond the design stage and considering the entire process from design to physical use. Connector C4 and similar cases underline the importance of spatial reference for DA frameworks that aim to include physical realization. Without contextual cues, users must mentally map the abstract 3D view onto the real bicycle, which is particularly difficult for users who are not accustomed to interpreting abstract CAD representations. Reference geometries, such as schematic handlebars, frame segments, or length scales, could help users position and orient connectors more reliably. In more advanced implementations, importing photos, CAD geometry, or 3D scans of the user’s bike could further improve spatial understanding. Together with improved measurement guidance and robust default dimensions for interfaces, such visual aids would likely increase first-shot success rates and reduce the need for design iterations.
5. Conclusion and outlook
This work presents an evaluation approach for DA framework through deliberately simple, accessible implementations and end-to-end assessment from user specification to physical testing. The approach is demonstrated on a modular DA framework that has been implemented in a Bike Connector Tool.
Implementing the framework in the Bike Connector Tool enabled the collection of empirical user feedback across the full end-to-end process chain. The conducted case study demonstrates that such simple, real-world implementations can yield meaningful framework-level insights. In particular, the results indicate that the framework architecture supports an efficient transition from design intent to production-ready geometry. At the same time, it highlights the importance of going beyond geometry generation and considering the entire end-to-end process, from design intent to the physical part.
Across participants of the case study, the tool was perceived as easy and intuitive to use, and most users indicated that the generated geometry matched their design intent. This shows that the framework enables quick progression from ideas to functional parts.
The case study further revealed limitations that directly inform future framework development. First, the perceived coverage of the feature library was insufficient in some scenarios, motivating a layered library strategy combining easy-to-use standard interfaces with more adaptable interfaces as well as allowing users to upload and share custom interfaces. Second, users requested the option to influence automated geometry generation when outcomes appeared unintuitive, indicating a need for optional manual control within automated steps. Third, the lack of spatial context in the UI often led to misjudged dimensions and orientations that only became apparent when mounting the physical parts, highlighting spatial reference and contextual visualization as critical elements for DA tools.
The evaluation relied on a qualitative case study and primarily reflects interactions of technically trained participants. The results confirm that users with CAD experience could use this tool to move from design to physical part very efficiently. However, thorough quantitative studies with controlled variables and measurable performance indicators are required to substantiate the observed trends, and additional studies are needed to understand how non-technical users interact with the framework, what guidance they require, and where failures occur. Moreover, the current demonstrator focuses on geometry generation; future extensions should incorporate additional aspects such as manufacturability feedback, build-orientation guidance, and structural evaluation, which become increasingly important when targeting less experienced users. Overall, the framework’s flexible architecture and the broad potential user base make the Bike Connector Tool a valuable testing ground for experimenting with new algorithmic modules, interaction patterns, or visualization strategies.
The Bike Connector Tool is publicly accessible and can be explored at: https://pdz.ethz.ch/research/dnt/bike-connector-tool.html
Open access
